> Might it be a consequence of lack of disk space in those early decades

A plausible explanation of "normalization as a process" was actually found in https://www.cargocultcode.com/normalization-is-not-a-process... ("So where did it begin?").

I hope someday to find some technical report of migrating to the relational database, from around that time.

> Normalization-as-process makes sense in a specific scenario: When converting a hierarchical database model into a relational model.

That makes much more sense as reasoning.

If I can also offer a second hazard of guess. I used to work in embedded in the 2000's and it was absolutely insane how almost all of the eldy architects and developers would readily accept some fixed width file format for data storage over a sensible solution that offered out of the box transactionality and relational modelling like Sqlite. This creates a mindset where each datastore is effectively siloed and must contain all the information to perform the operation, potentially leading to these denormalised designs.

Bit weird, given that was from the waterfall era, implying that the "Big Design Up Front" wasn't actually doing any real thinking about modelling up front. But I've been in that room and I think a lot of it was cargo cult. To deal with the insanity of simple file I/O as data, I had to write a rudimentary atomicity system from scratch in order to fix the dumb corruption issues of their design when I would have got that for free with Sqlite.